WP/07/141
Testing for Cointegration Using the Johansen Methodology when Variables are Near-Integrated Erik Hjalmarsson and Pär Österholm
© 2007 International Monetary Fund
WP/07/141
IMF Working Paper Western Hemisphere Division Testing for Cointegration Using the Johansen Methodology when Variables are NearIntegrated Prepared by Erik Hjalmarsson and Pär Österholm∗ Authorized for distribution by Robert K. Rennhack June 2007 Abstract This Working Paper should not be reported as representing the views of the IMF. The views expressed in this Working Paper are those of the author(s) and do not necessarily represent those of the IMF or IMF policy. Working Papers describe research in progress by the author(s) and are published to elicit comments and to further debate.
We investigate the properties of Johansen’s (1988, 1991) maximum eigenvalue and trace tests for cointegration under the empirically relevant situation of near-integrated variables. Using Monte Carlo techniques, we show that in a system with near-integrated variables, the probability of reaching an erroneous conclusion regarding the cointegrating rank of the system is generally substantially higher than the nominal size. The risk of concluding that completely unrelated series are cointegrated is therefore non-negligible. The spurious rejection rate can be reduced by performing additional tests of restrictions on the cointegrating vector(s), although it is still substantially larger than the nominal size. JEL Classification Numbers: C12, C15, C32 Keywords: Cointegration, Near-unit-roots, Spurious rejection, Monte Carlo simulations Authors’ E-Mail Addresses: ∗
[email protected];
[email protected]
Hjalmarsson: Board of Governors of the Federal Reserve System. Österholm: IMF and Department of Economics, Uppsala University. We are grateful to Meredith Beechey, Lennart Hjalmarsson and Randi Hjalmarsson for valuable comments on this paper and to Benjamin Chiquoine for excellent research assistance. Österholm gratefully acknowledges financial support from Jan Wallander’s and Tom Hedelius’ foundation. The views in this paper are solely the responsibility of the authors and should not be interpreted as reflecting the views of the Board of Governors the Federal Reserve System or of any other person associated with the Federal Reserve System.
2 Contents
Page
I. Introduction ...........................................................................................................................3 II. Testing for Cointegration Using Johansen’s Methodology..................................................4 III. Monte Carlo Study..............................................................................................................7 A. Setup......................................................................................................................7 B. Results ...................................................................................................................9 IV. An Empirical Illustration ..................................................................................................12 V. Conclusion .........................................................................................................................15 Appendix...................................................................................................................................17 References.................................................................................................................................18 Tables 1. Results from Augmented Dickey-Fuller Test .....................................................................13 2. Results from Cointegration Test .........................................................................................14 3. Results from Hypothesis Tests on the Cointegrating Vector..............................................14 Figures 1. Spurious Rejection Frequency for Bivariate System ..........................................................10 2. Spurious Rejection Frequency for Trivariate System.........................................................11 3. Data .....................................................................................................................................13
3 I. INTRODUCTION Cointegration methods have been very popular tools in applied economic work since their introduction about twenty years ago. However, the strict unit-root assumption that these methods typically rely upon is often not easy to justify on economic or theoretical grounds. For instance, variables such as inflation, interest rates, real exchange rates and unemployment rates all appear to be highly persistent, and are frequently modelled as unit root processes. But, there is little a priori reason to believe that these variables have an exact unit root, rather than a root close to unity. In fact, these variables often show signs of mean reversion in long enough samples.1 Since unit-root tests have very limited power to distinguish between a unit-root and a close alternative, the pure unit-root assumption is typically based on convenience rather than on strong theoretical or empirical facts. This has led many economists and econometricians to believe near-integrated processes, which explicitly allow for a small (unknown) deviation from the pure unit-root assumption, to be a more appropriate way to describe many economic time series; see, for example, Stock (1991), Cavanagh et al., (1995) and Elliott (1998).2 Near-integrated and integrated time series have implications for estimation and inference that are similar in many respects. For instance, spurious regressions are a problem when variables are near-integrated as well as integrated, and therefore, it is also relevant to discuss cointegration of near-integrated variables; see Phillips (1988) for an analytical discussion regarding these issues. Unfortunately, inferential procedures designed for data generated by unit-root processes tend not to be robust to deviations from the unit-root assumption. For instance, Elliott (1998) shows that large size distortions can occur when performing inference on the cointegration vector in a system where the individual variables follow near-unit-root processes rather than pure unit-root processes. The purpose of this paper is to investigate the effect of deviations from the unit-root assumption on the determination of the cointegrating rank of the system using Johansen’s (1988, 1991) maximum eigenvalue and trace tests. Unlike inference regarding the cointegrating vectors, this issue has not been investigated much in the literature. The first contribution of the current paper is therefore to document the rejection rates for standard tests of cointegration, using the Johansen framework, in a system where the variables are nearintegrated. Through extensive Monte Carlo simulations, we show that the probability of 1
For studies relying on cointegration methods, see, for instance, Wallace and Warner (1993), Malley and Moutos (1996), Cardoso (1998), Bremnes et al. (2001), Jonsson (2001), Khamis and Leone (2001) and Bagchi et al. (2004). Studies arguing the stationarity of these variables include Song and Wu (1997, 1998), Taylor and Sarno (1998), Wu and Chen (2001) and Basher and Westerlund (2006). 2
Phillips (1988) considers both processes that have roots smaller than unity (“strongly autoregressive”) and larger than unity (“mildly explosive”) in his analysis of near-integrated processes. In this paper, however, we only consider the empirically most relevant case of processes with roots less than unity.
4 reaching an erroneous conclusion regarding the cointegrating rank of the system is generally substantially higher than the nominal size. That is, the nominal size of the test can vastly understate the risks of finding a spurious relationship between unrelated near-integrated variables. In a simple bivariate system, the spurious rejection rate can approach 20 and 40 percent for the maximum eigenvalue and trace tests respectively, using a nominal size of five percent. Even higher rejection rates are found in a trivariate system. The second contribution is to show how a sequence of additional tests on the cointegrating vector(s) can help improve the performance of the tests and reduce the spurious rejection rate. However, even after taking these extra steps, the rejection rate of the test is still considerably larger than the nominal size. This is particularly true for the trivariate system where spurious rejection rates between 15 and 20 percent are documented for nominal five percent tests. Overall, the performance of the trace test appears worse than that of the maximum eigenvalue test. Both tests, however, have large enough deviations from the nominal size that practitioners should be aware of the problems associated with Johansen’s procedures under these circumstances. The proposed sequence of additional tests helps alleviate some of the sensitivity of the Johansen procedures to deviations from the strict unit-root assumption. They do not, however, eliminate the problem. The remainder of this paper is organised as follows: Section II gives a brief introduction to Johansen’s methodology and Section III presents the Monte Carlo study. In Section IV, we present an empirical illustration of the problems associated with near-integrated variables using U.S. data on CPI inflation and the short nominal interest rate. Section V concludes. II. TESTING FOR COINTEGRATION USING JOHANSEN’S METHODOLOGY Johansen’s methodology takes its starting point in the vector autoregression (VAR) of order p given by y t = μ + A 1 y t −1 + " + A p y t − p + ε t ,
(1)
where y t is an nx1 vector of variables that are integrated of order one – commonly denoted I(1) – and ε t is an nx1 vector of innovations. This VAR can be re-written as p −1
Δy t = μ + Πy t −1 + ∑ Γ i Δy t −i + ε t
(2)
i =1
where p
p
i =1
j = i +1
Π = ∑ A i − I and Γ i = − ∑ A j .
(3)
5 If the coefficient matrix Π has reduced rank r
